Details have emerged about NVIDIA’s next-generation consumer GPU architecture codenamed “Blackwell,” expected to power the future RTX 50 series graphics cards in 2025, leaks flagship RTX 5090 is slated to be powered by NVIDIA’s GB202 GPU supporting ultra-fast GDDR7 memory.
The GB202 is said to utilize a 384-bit memory bus paired with cutting-edge GDDR7 RAM from Micron and Samsung, with individual GDDR7 chips reaching speeds up to 32Gbps, this adds up to a massive 1536GB/s memory bandwidth on the RTX 5090.
GDDR7 also enables higher capacity up to 24Gb per chip, combination of extreme speed and capacity will feed the Blackwell GPU for next-level gaming performance.
NVIDIA’s second-tier RTX 5080 card is rumored to use the slightly scaled back GB203 GPU which will support a 256-bit GDDR7 memory bus, while a step below the top-end RTX 5090, memory bandwidth will still see a huge uplift versus current GDDR6X.
Leaks suggest the flagship RTX 5090 reference card will resemble the RTX 4090 Ti exposed design with a bulky 4-slot form factor and side-mounted PCB. Cooling will need to keep pace with the ultra-powerful Blackwell GPU and super-fast GDDR7.
AMD also has its next-gen RDNA5 GPU architecture in the pipeline to compete with Nvidia RTX 50 series, flagship RX 9900 XTX graphics cards are expected around 6 months after the RTX 50 launch, directly taking on the RTX 5090.
NVIDIA’s Blackwell architecture and GDDR7 represent the bleeding edge of graphics performance that will power immersive 8K gaming and next-gen creative workloads. For PC enthusiasts, the RTX 50 series will sit at the pinnacle of visual fidelity, ray tracing, and speed, availability in 2025 once GDDR7 volume production is smoothed out.